A learner can understand every instruction about tongue position and still hear two English sounds as one category. When that happens, another ten repetitions may rehearse the same guess. Pronunciation lessons should often begin with listening discrimination, then move into production and return to listening. The ear gives the mouth a target.
This claim is narrower than “listen more.” The learner needs brief, answerable trials with feedback: Which word did you hear? Are these two tokens the same or different? Which syllable carries the main stress? Passive exposure to a podcast cannot show whether the relevant contrast was detected.
Hearing supplies the target
Speech categories are learned through distributions, context, and experience. An English contrast may divide sound space differently from華語. A Taiwanese learner who has difficulty with /l/ and /r/, final voicing, or reduced vowels may know the spelling and the articulatory explanation while perception remains unstable.
In a study of 115 Japanese university learners of English, Lee, Jang, and Plonsky compared perception-based and production-based pronunciation instruction. Students received two weeks of work on either segmental or suprasegmental features. All four treatment groups improved, with patterns varying by feature and test time. The authors judged perception-based training the stronger option across the two feature types. The design supports a practical preference, though Japanese learners and a two-week course do not settle the sequence for every Taiwanese class.
The finding also resists a familiar classroom shortcut. A learner who can imitate one model immediately may be using a short acoustic trace. Test the contrast again with a different word, a different speaker, and a short delay.
Build a small listening block
Choose one contrast that carries a real cost for this learner. For /l/ and /r/, a six-minute block might contain 18 trials:
- six same-or-different pairs, including two identical pairs,
- eight word identifications spoken by two voices,
- four short phrases in which the target word changes the message.
Do not display the spelling on the first pass. Orthography can pull attention toward a familiar written category. After each choice, give the correct answer and replay once. Ask for confidence on a three-point scale. A correct low-confidence response should remain in the practice set.
Numbers matter here. If a learner identifies 7 of 12 items before production and 10 of 12 afterward, the lesson has more information than “sounds better.” Keep novel voices for the final four trials. Choc's article on changing the voice during ear training explains why success with one speaker can stay tied to that speaker.
Move from choice to movement
Once the learner can hear the target above chance, connect the sound to an articulatory action. Use a mirror, a side-view diagram, or a short video. For /r/, the tongue shape varies across speakers, so an instruction that demands one exact posture can be misleading. Focus on the acoustic result and a workable movement for this learner.
Production can follow a progression from syllable to word, phrase, and a sentence whose meaning requires the target. Record three attempts. The learner chooses the clearest one before receiving teacher feedback. That choice forces another perception judgment.
The route also runs in the other direction. A 2024 experiment with English-speaking learners of German found that explicit production training improved discrimination accuracy by 19% and identification by 14%. The targets were German vowels and fricatives, so those percentages should stay attached to that sample and task. They do show that mouth work can sharpen listening, especially when articulation is taught explicitly.
That is why a strict listening-only rule would be too simple. Perception and production can support each other. Sequencing them gives the teacher a way to locate the blockage.
Return to unfamiliar speech
End the lesson with new tokens rather than a polished repetition of the opening list. Change one variable at a time. Use the same words with a new voice, then new words with the familiar voice. If both change together and accuracy falls, the teacher cannot tell which change caused the trouble.
High-variability phonetic training often mixes speakers and contexts so learners do not memorize one recording. A 2021 systematic review and meta-analysis of talker variability found a small immediate summary effect that shrank after correction for publication bias, alongside larger estimates for transfer to new talkers and delayed retention. The review stressed heterogeneity and research limits. Training gains do not guarantee spontaneous use in conversation.
Feedback must identify the decision. A red mark beside a minimal pair says that one response failed, while a replay with the two alternatives tells the learner where the boundary lies. After a wrong answer, play the target beside the closest competitor, ask again, then insert a different voice. That sequence takes roughly 23 seconds and supplies another retrieval attempt.
Avoid chasing perfection on every token. A contrast deserves more class time when confusion changes the word, obscures a message, or repeatedly slows the listener. A rare accent feature with little effect on comprehensibility can wait. I would rather finish with nine stable identifications across three voices than 40 polished repetitions copied from one recording.
Delayed checks are useful. Test 11 items two days later, before another explanation. Include five trained words, four new words, and two catch items. A learner may retain identification while production slips, or the reverse. Those patterns call for different homework.
Cases where production can start first
Some learners already identify a contrast accurately and lack motor control. Their first block should be production. An actor preparing a line, a singer matching vowel length, or an advanced speaker adjusting prominence may gain little from elementary same-or-different trials.
Age, hearing, attention, and the target feature also change the plan. Rhythm work may need tapping and movement early. A learner with suspected hearing loss needs an audiological route, not a longer minimal-pair worksheet.
The teacher should test before choosing. Twelve perception items and three recorded productions take under eight minutes. If listening accuracy is 11 out of 12 and production is consistently unclear, move to the mouth. If listening sits at 6 out of 12, keep the first practice block in the headphones.
Tomorrow's lesson can end where it began: play four unseen items without text and ask for a choice plus confidence. The final answer is useful. The hesitation before it is useful too.